Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
58862699 | 2 | * linux/fs/jbd2/revoke.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 2000 | |
5 | * | |
6 | * Copyright 2000 Red Hat corp --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Journal revoke routines for the generic filesystem journaling code; | |
13 | * part of the ext2fs journaling system. | |
14 | * | |
15 | * Revoke is the mechanism used to prevent old log records for deleted | |
16 | * metadata from being replayed on top of newer data using the same | |
17 | * blocks. The revoke mechanism is used in two separate places: | |
18 | * | |
19 | * + Commit: during commit we write the entire list of the current | |
20 | * transaction's revoked blocks to the journal | |
21 | * | |
22 | * + Recovery: during recovery we record the transaction ID of all | |
23 | * revoked blocks. If there are multiple revoke records in the log | |
24 | * for a single block, only the last one counts, and if there is a log | |
25 | * entry for a block beyond the last revoke, then that log entry still | |
26 | * gets replayed. | |
27 | * | |
28 | * We can get interactions between revokes and new log data within a | |
29 | * single transaction: | |
30 | * | |
31 | * Block is revoked and then journaled: | |
32 | * The desired end result is the journaling of the new block, so we | |
33 | * cancel the revoke before the transaction commits. | |
34 | * | |
35 | * Block is journaled and then revoked: | |
36 | * The revoke must take precedence over the write of the block, so we | |
37 | * need either to cancel the journal entry or to write the revoke | |
38 | * later in the log than the log block. In this case, we choose the | |
39 | * latter: journaling a block cancels any revoke record for that block | |
40 | * in the current transaction, so any revoke for that block in the | |
41 | * transaction must have happened after the block was journaled and so | |
42 | * the revoke must take precedence. | |
43 | * | |
44 | * Block is revoked and then written as data: | |
45 | * The data write is allowed to succeed, but the revoke is _not_ | |
46 | * cancelled. We still need to prevent old log records from | |
47 | * overwriting the new data. We don't even need to clear the revoke | |
48 | * bit here. | |
49 | * | |
50 | * Revoke information on buffers is a tri-state value: | |
51 | * | |
52 | * RevokeValid clear: no cached revoke status, need to look it up | |
53 | * RevokeValid set, Revoked clear: | |
54 | * buffer has not been revoked, and cancel_revoke | |
55 | * need do nothing. | |
56 | * RevokeValid set, Revoked set: | |
57 | * buffer has been revoked. | |
86db97c8 JK |
58 | * |
59 | * Locking rules: | |
60 | * We keep two hash tables of revoke records. One hashtable belongs to the | |
61 | * running transaction (is pointed to by journal->j_revoke), the other one | |
62 | * belongs to the committing transaction. Accesses to the second hash table | |
63 | * happen only from the kjournald and no other thread touches this table. Also | |
64 | * journal_switch_revoke_table() which switches which hashtable belongs to the | |
65 | * running and which to the committing transaction is called only from | |
66 | * kjournald. Therefore we need no locks when accessing the hashtable belonging | |
67 | * to the committing transaction. | |
68 | * | |
69 | * All users operating on the hash table belonging to the running transaction | |
70 | * have a handle to the transaction. Therefore they are safe from kjournald | |
71 | * switching hash tables under them. For operations on the lists of entries in | |
72 | * the hash table j_revoke_lock is used. | |
73 | * | |
74 | * Finally, also replay code uses the hash tables but at this moment noone else | |
75 | * can touch them (filesystem isn't mounted yet) and hence no locking is | |
76 | * needed. | |
470decc6 DK |
77 | */ |
78 | ||
79 | #ifndef __KERNEL__ | |
80 | #include "jfs_user.h" | |
81 | #else | |
82 | #include <linux/time.h> | |
83 | #include <linux/fs.h> | |
f7f4bccb | 84 | #include <linux/jbd2.h> |
470decc6 DK |
85 | #include <linux/errno.h> |
86 | #include <linux/slab.h> | |
87 | #include <linux/list.h> | |
470decc6 DK |
88 | #include <linux/init.h> |
89 | #endif | |
f482394c | 90 | #include <linux/log2.h> |
470decc6 | 91 | |
e18b890b CL |
92 | static struct kmem_cache *jbd2_revoke_record_cache; |
93 | static struct kmem_cache *jbd2_revoke_table_cache; | |
470decc6 DK |
94 | |
95 | /* Each revoke record represents one single revoked block. During | |
96 | journal replay, this involves recording the transaction ID of the | |
97 | last transaction to revoke this block. */ | |
98 | ||
f7f4bccb | 99 | struct jbd2_revoke_record_s |
470decc6 DK |
100 | { |
101 | struct list_head hash; | |
102 | tid_t sequence; /* Used for recovery only */ | |
18eba7aa | 103 | unsigned long long blocknr; |
470decc6 DK |
104 | }; |
105 | ||
106 | ||
107 | /* The revoke table is just a simple hash table of revoke records. */ | |
f7f4bccb | 108 | struct jbd2_revoke_table_s |
470decc6 DK |
109 | { |
110 | /* It is conceivable that we might want a larger hash table | |
111 | * for recovery. Must be a power of two. */ | |
112 | int hash_size; | |
113 | int hash_shift; | |
114 | struct list_head *hash_table; | |
115 | }; | |
116 | ||
117 | ||
118 | #ifdef __KERNEL__ | |
119 | static void write_one_revoke_record(journal_t *, transaction_t *, | |
120 | struct journal_head **, int *, | |
f7f4bccb | 121 | struct jbd2_revoke_record_s *); |
470decc6 DK |
122 | static void flush_descriptor(journal_t *, struct journal_head *, int); |
123 | #endif | |
124 | ||
125 | /* Utility functions to maintain the revoke table */ | |
126 | ||
127 | /* Borrowed from buffer.c: this is a tried and tested block hash function */ | |
18eba7aa | 128 | static inline int hash(journal_t *journal, unsigned long long block) |
470decc6 | 129 | { |
f7f4bccb | 130 | struct jbd2_revoke_table_s *table = journal->j_revoke; |
470decc6 | 131 | int hash_shift = table->hash_shift; |
29971769 | 132 | int hash = (int)block ^ (int)((block >> 31) >> 1); |
470decc6 | 133 | |
29971769 MC |
134 | return ((hash << (hash_shift - 6)) ^ |
135 | (hash >> 13) ^ | |
136 | (hash << (hash_shift - 12))) & (table->hash_size - 1); | |
470decc6 DK |
137 | } |
138 | ||
18eba7aa | 139 | static int insert_revoke_hash(journal_t *journal, unsigned long long blocknr, |
470decc6 DK |
140 | tid_t seq) |
141 | { | |
142 | struct list_head *hash_list; | |
f7f4bccb | 143 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
144 | |
145 | repeat: | |
f7f4bccb | 146 | record = kmem_cache_alloc(jbd2_revoke_record_cache, GFP_NOFS); |
470decc6 DK |
147 | if (!record) |
148 | goto oom; | |
149 | ||
150 | record->sequence = seq; | |
151 | record->blocknr = blocknr; | |
152 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
153 | spin_lock(&journal->j_revoke_lock); | |
154 | list_add(&record->hash, hash_list); | |
155 | spin_unlock(&journal->j_revoke_lock); | |
156 | return 0; | |
157 | ||
158 | oom: | |
159 | if (!journal_oom_retry) | |
160 | return -ENOMEM; | |
329d291f | 161 | jbd_debug(1, "ENOMEM in %s, retrying\n", __func__); |
470decc6 DK |
162 | yield(); |
163 | goto repeat; | |
164 | } | |
165 | ||
166 | /* Find a revoke record in the journal's hash table. */ | |
167 | ||
f7f4bccb | 168 | static struct jbd2_revoke_record_s *find_revoke_record(journal_t *journal, |
18eba7aa | 169 | unsigned long long blocknr) |
470decc6 DK |
170 | { |
171 | struct list_head *hash_list; | |
f7f4bccb | 172 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
173 | |
174 | hash_list = &journal->j_revoke->hash_table[hash(journal, blocknr)]; | |
175 | ||
176 | spin_lock(&journal->j_revoke_lock); | |
f7f4bccb | 177 | record = (struct jbd2_revoke_record_s *) hash_list->next; |
470decc6 DK |
178 | while (&(record->hash) != hash_list) { |
179 | if (record->blocknr == blocknr) { | |
180 | spin_unlock(&journal->j_revoke_lock); | |
181 | return record; | |
182 | } | |
f7f4bccb | 183 | record = (struct jbd2_revoke_record_s *) record->hash.next; |
470decc6 DK |
184 | } |
185 | spin_unlock(&journal->j_revoke_lock); | |
186 | return NULL; | |
187 | } | |
188 | ||
9fa27c85 DG |
189 | void jbd2_journal_destroy_revoke_caches(void) |
190 | { | |
191 | if (jbd2_revoke_record_cache) { | |
192 | kmem_cache_destroy(jbd2_revoke_record_cache); | |
193 | jbd2_revoke_record_cache = NULL; | |
194 | } | |
195 | if (jbd2_revoke_table_cache) { | |
196 | kmem_cache_destroy(jbd2_revoke_table_cache); | |
197 | jbd2_revoke_table_cache = NULL; | |
198 | } | |
199 | } | |
200 | ||
f7f4bccb | 201 | int __init jbd2_journal_init_revoke_caches(void) |
470decc6 | 202 | { |
9fa27c85 DG |
203 | J_ASSERT(!jbd2_revoke_record_cache); |
204 | J_ASSERT(!jbd2_revoke_table_cache); | |
205 | ||
a920e941 | 206 | jbd2_revoke_record_cache = kmem_cache_create("jbd2_revoke_record", |
f7f4bccb | 207 | sizeof(struct jbd2_revoke_record_s), |
77160957 MC |
208 | 0, |
209 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
210 | NULL); | |
1076d17a | 211 | if (!jbd2_revoke_record_cache) |
9fa27c85 | 212 | goto record_cache_failure; |
470decc6 | 213 | |
a920e941 | 214 | jbd2_revoke_table_cache = kmem_cache_create("jbd2_revoke_table", |
f7f4bccb | 215 | sizeof(struct jbd2_revoke_table_s), |
77160957 | 216 | 0, SLAB_TEMPORARY, NULL); |
9fa27c85 DG |
217 | if (!jbd2_revoke_table_cache) |
218 | goto table_cache_failure; | |
470decc6 | 219 | return 0; |
9fa27c85 DG |
220 | table_cache_failure: |
221 | jbd2_journal_destroy_revoke_caches(); | |
222 | record_cache_failure: | |
223 | return -ENOMEM; | |
470decc6 DK |
224 | } |
225 | ||
83c49523 | 226 | static struct jbd2_revoke_table_s *jbd2_journal_init_revoke_table(int hash_size) |
470decc6 | 227 | { |
83c49523 DG |
228 | int shift = 0; |
229 | int tmp = hash_size; | |
230 | struct jbd2_revoke_table_s *table; | |
470decc6 | 231 | |
83c49523 DG |
232 | table = kmem_cache_alloc(jbd2_revoke_table_cache, GFP_KERNEL); |
233 | if (!table) | |
234 | goto out; | |
470decc6 | 235 | |
470decc6 DK |
236 | while((tmp >>= 1UL) != 0UL) |
237 | shift++; | |
238 | ||
83c49523 DG |
239 | table->hash_size = hash_size; |
240 | table->hash_shift = shift; | |
241 | table->hash_table = | |
470decc6 | 242 | kmalloc(hash_size * sizeof(struct list_head), GFP_KERNEL); |
83c49523 DG |
243 | if (!table->hash_table) { |
244 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
245 | table = NULL; | |
246 | goto out; | |
470decc6 DK |
247 | } |
248 | ||
249 | for (tmp = 0; tmp < hash_size; tmp++) | |
83c49523 | 250 | INIT_LIST_HEAD(&table->hash_table[tmp]); |
470decc6 | 251 | |
83c49523 DG |
252 | out: |
253 | return table; | |
254 | } | |
255 | ||
256 | static void jbd2_journal_destroy_revoke_table(struct jbd2_revoke_table_s *table) | |
257 | { | |
258 | int i; | |
259 | struct list_head *hash_list; | |
260 | ||
261 | for (i = 0; i < table->hash_size; i++) { | |
262 | hash_list = &table->hash_table[i]; | |
263 | J_ASSERT(list_empty(hash_list)); | |
470decc6 DK |
264 | } |
265 | ||
83c49523 DG |
266 | kfree(table->hash_table); |
267 | kmem_cache_free(jbd2_revoke_table_cache, table); | |
268 | } | |
470decc6 | 269 | |
83c49523 DG |
270 | /* Initialise the revoke table for a given journal to a given size. */ |
271 | int jbd2_journal_init_revoke(journal_t *journal, int hash_size) | |
272 | { | |
273 | J_ASSERT(journal->j_revoke_table[0] == NULL); | |
f482394c | 274 | J_ASSERT(is_power_of_2(hash_size)); |
470decc6 | 275 | |
83c49523 DG |
276 | journal->j_revoke_table[0] = jbd2_journal_init_revoke_table(hash_size); |
277 | if (!journal->j_revoke_table[0]) | |
278 | goto fail0; | |
470decc6 | 279 | |
83c49523 DG |
280 | journal->j_revoke_table[1] = jbd2_journal_init_revoke_table(hash_size); |
281 | if (!journal->j_revoke_table[1]) | |
282 | goto fail1; | |
470decc6 | 283 | |
83c49523 | 284 | journal->j_revoke = journal->j_revoke_table[1]; |
470decc6 DK |
285 | |
286 | spin_lock_init(&journal->j_revoke_lock); | |
287 | ||
288 | return 0; | |
470decc6 | 289 | |
83c49523 DG |
290 | fail1: |
291 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
292 | fail0: | |
293 | return -ENOMEM; | |
294 | } | |
470decc6 | 295 | |
83c49523 | 296 | /* Destroy a journal's revoke table. The table must already be empty! */ |
f7f4bccb | 297 | void jbd2_journal_destroy_revoke(journal_t *journal) |
470decc6 | 298 | { |
470decc6 | 299 | journal->j_revoke = NULL; |
83c49523 DG |
300 | if (journal->j_revoke_table[0]) |
301 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[0]); | |
302 | if (journal->j_revoke_table[1]) | |
303 | jbd2_journal_destroy_revoke_table(journal->j_revoke_table[1]); | |
470decc6 DK |
304 | } |
305 | ||
306 | ||
307 | #ifdef __KERNEL__ | |
308 | ||
309 | /* | |
f7f4bccb | 310 | * jbd2_journal_revoke: revoke a given buffer_head from the journal. This |
470decc6 DK |
311 | * prevents the block from being replayed during recovery if we take a |
312 | * crash after this current transaction commits. Any subsequent | |
313 | * metadata writes of the buffer in this transaction cancel the | |
314 | * revoke. | |
315 | * | |
316 | * Note that this call may block --- it is up to the caller to make | |
317 | * sure that there are no further calls to journal_write_metadata | |
318 | * before the revoke is complete. In ext3, this implies calling the | |
319 | * revoke before clearing the block bitmap when we are deleting | |
320 | * metadata. | |
321 | * | |
f7f4bccb | 322 | * Revoke performs a jbd2_journal_forget on any buffer_head passed in as a |
470decc6 DK |
323 | * parameter, but does _not_ forget the buffer_head if the bh was only |
324 | * found implicitly. | |
325 | * | |
326 | * bh_in may not be a journalled buffer - it may have come off | |
327 | * the hash tables without an attached journal_head. | |
328 | * | |
f7f4bccb | 329 | * If bh_in is non-zero, jbd2_journal_revoke() will decrement its b_count |
470decc6 DK |
330 | * by one. |
331 | */ | |
332 | ||
18eba7aa | 333 | int jbd2_journal_revoke(handle_t *handle, unsigned long long blocknr, |
470decc6 DK |
334 | struct buffer_head *bh_in) |
335 | { | |
336 | struct buffer_head *bh = NULL; | |
337 | journal_t *journal; | |
338 | struct block_device *bdev; | |
339 | int err; | |
340 | ||
341 | might_sleep(); | |
342 | if (bh_in) | |
343 | BUFFER_TRACE(bh_in, "enter"); | |
344 | ||
345 | journal = handle->h_transaction->t_journal; | |
f7f4bccb | 346 | if (!jbd2_journal_set_features(journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)){ |
470decc6 DK |
347 | J_ASSERT (!"Cannot set revoke feature!"); |
348 | return -EINVAL; | |
349 | } | |
350 | ||
351 | bdev = journal->j_fs_dev; | |
352 | bh = bh_in; | |
353 | ||
354 | if (!bh) { | |
355 | bh = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
356 | if (bh) | |
357 | BUFFER_TRACE(bh, "found on hash"); | |
358 | } | |
cd02ff0b | 359 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
360 | else { |
361 | struct buffer_head *bh2; | |
362 | ||
363 | /* If there is a different buffer_head lying around in | |
364 | * memory anywhere... */ | |
365 | bh2 = __find_get_block(bdev, blocknr, journal->j_blocksize); | |
366 | if (bh2) { | |
367 | /* ... and it has RevokeValid status... */ | |
368 | if (bh2 != bh && buffer_revokevalid(bh2)) | |
369 | /* ...then it better be revoked too, | |
370 | * since it's illegal to create a revoke | |
371 | * record against a buffer_head which is | |
372 | * not marked revoked --- that would | |
373 | * risk missing a subsequent revoke | |
374 | * cancel. */ | |
375 | J_ASSERT_BH(bh2, buffer_revoked(bh2)); | |
376 | put_bh(bh2); | |
377 | } | |
378 | } | |
379 | #endif | |
380 | ||
381 | /* We really ought not ever to revoke twice in a row without | |
382 | first having the revoke cancelled: it's illegal to free a | |
383 | block twice without allocating it in between! */ | |
384 | if (bh) { | |
385 | if (!J_EXPECT_BH(bh, !buffer_revoked(bh), | |
386 | "inconsistent data on disk")) { | |
387 | if (!bh_in) | |
388 | brelse(bh); | |
389 | return -EIO; | |
390 | } | |
391 | set_buffer_revoked(bh); | |
392 | set_buffer_revokevalid(bh); | |
393 | if (bh_in) { | |
f7f4bccb MC |
394 | BUFFER_TRACE(bh_in, "call jbd2_journal_forget"); |
395 | jbd2_journal_forget(handle, bh_in); | |
470decc6 DK |
396 | } else { |
397 | BUFFER_TRACE(bh, "call brelse"); | |
398 | __brelse(bh); | |
399 | } | |
400 | } | |
401 | ||
29971769 | 402 | jbd_debug(2, "insert revoke for block %llu, bh_in=%p\n",blocknr, bh_in); |
470decc6 DK |
403 | err = insert_revoke_hash(journal, blocknr, |
404 | handle->h_transaction->t_tid); | |
405 | BUFFER_TRACE(bh_in, "exit"); | |
406 | return err; | |
407 | } | |
408 | ||
409 | /* | |
410 | * Cancel an outstanding revoke. For use only internally by the | |
f7f4bccb | 411 | * journaling code (called from jbd2_journal_get_write_access). |
470decc6 DK |
412 | * |
413 | * We trust buffer_revoked() on the buffer if the buffer is already | |
414 | * being journaled: if there is no revoke pending on the buffer, then we | |
415 | * don't do anything here. | |
416 | * | |
417 | * This would break if it were possible for a buffer to be revoked and | |
418 | * discarded, and then reallocated within the same transaction. In such | |
419 | * a case we would have lost the revoked bit, but when we arrived here | |
420 | * the second time we would still have a pending revoke to cancel. So, | |
421 | * do not trust the Revoked bit on buffers unless RevokeValid is also | |
422 | * set. | |
470decc6 | 423 | */ |
f7f4bccb | 424 | int jbd2_journal_cancel_revoke(handle_t *handle, struct journal_head *jh) |
470decc6 | 425 | { |
f7f4bccb | 426 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
427 | journal_t *journal = handle->h_transaction->t_journal; |
428 | int need_cancel; | |
429 | int did_revoke = 0; /* akpm: debug */ | |
430 | struct buffer_head *bh = jh2bh(jh); | |
431 | ||
432 | jbd_debug(4, "journal_head %p, cancelling revoke\n", jh); | |
433 | ||
434 | /* Is the existing Revoke bit valid? If so, we trust it, and | |
435 | * only perform the full cancel if the revoke bit is set. If | |
436 | * not, we can't trust the revoke bit, and we need to do the | |
437 | * full search for a revoke record. */ | |
438 | if (test_set_buffer_revokevalid(bh)) { | |
439 | need_cancel = test_clear_buffer_revoked(bh); | |
440 | } else { | |
441 | need_cancel = 1; | |
442 | clear_buffer_revoked(bh); | |
443 | } | |
444 | ||
445 | if (need_cancel) { | |
446 | record = find_revoke_record(journal, bh->b_blocknr); | |
447 | if (record) { | |
448 | jbd_debug(4, "cancelled existing revoke on " | |
449 | "blocknr %llu\n", (unsigned long long)bh->b_blocknr); | |
450 | spin_lock(&journal->j_revoke_lock); | |
451 | list_del(&record->hash); | |
452 | spin_unlock(&journal->j_revoke_lock); | |
f7f4bccb | 453 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
454 | did_revoke = 1; |
455 | } | |
456 | } | |
457 | ||
cd02ff0b | 458 | #ifdef JBD2_EXPENSIVE_CHECKING |
470decc6 DK |
459 | /* There better not be one left behind by now! */ |
460 | record = find_revoke_record(journal, bh->b_blocknr); | |
461 | J_ASSERT_JH(jh, record == NULL); | |
462 | #endif | |
463 | ||
464 | /* Finally, have we just cleared revoke on an unhashed | |
465 | * buffer_head? If so, we'd better make sure we clear the | |
466 | * revoked status on any hashed alias too, otherwise the revoke | |
467 | * state machine will get very upset later on. */ | |
468 | if (need_cancel) { | |
469 | struct buffer_head *bh2; | |
470 | bh2 = __find_get_block(bh->b_bdev, bh->b_blocknr, bh->b_size); | |
471 | if (bh2) { | |
472 | if (bh2 != bh) | |
473 | clear_buffer_revoked(bh2); | |
474 | __brelse(bh2); | |
475 | } | |
476 | } | |
477 | return did_revoke; | |
478 | } | |
479 | ||
480 | /* journal_switch_revoke table select j_revoke for next transaction | |
481 | * we do not want to suspend any processing until all revokes are | |
482 | * written -bzzz | |
483 | */ | |
f7f4bccb | 484 | void jbd2_journal_switch_revoke_table(journal_t *journal) |
470decc6 DK |
485 | { |
486 | int i; | |
487 | ||
488 | if (journal->j_revoke == journal->j_revoke_table[0]) | |
489 | journal->j_revoke = journal->j_revoke_table[1]; | |
490 | else | |
491 | journal->j_revoke = journal->j_revoke_table[0]; | |
492 | ||
493 | for (i = 0; i < journal->j_revoke->hash_size; i++) | |
494 | INIT_LIST_HEAD(&journal->j_revoke->hash_table[i]); | |
495 | } | |
496 | ||
497 | /* | |
498 | * Write revoke records to the journal for all entries in the current | |
499 | * revoke hash, deleting the entries as we go. | |
470decc6 | 500 | */ |
f7f4bccb | 501 | void jbd2_journal_write_revoke_records(journal_t *journal, |
470decc6 DK |
502 | transaction_t *transaction) |
503 | { | |
504 | struct journal_head *descriptor; | |
f7f4bccb MC |
505 | struct jbd2_revoke_record_s *record; |
506 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
507 | struct list_head *hash_list; |
508 | int i, offset, count; | |
509 | ||
510 | descriptor = NULL; | |
511 | offset = 0; | |
512 | count = 0; | |
513 | ||
514 | /* select revoke table for committing transaction */ | |
515 | revoke = journal->j_revoke == journal->j_revoke_table[0] ? | |
516 | journal->j_revoke_table[1] : journal->j_revoke_table[0]; | |
517 | ||
518 | for (i = 0; i < revoke->hash_size; i++) { | |
519 | hash_list = &revoke->hash_table[i]; | |
520 | ||
521 | while (!list_empty(hash_list)) { | |
f7f4bccb | 522 | record = (struct jbd2_revoke_record_s *) |
470decc6 DK |
523 | hash_list->next; |
524 | write_one_revoke_record(journal, transaction, | |
525 | &descriptor, &offset, | |
526 | record); | |
527 | count++; | |
528 | list_del(&record->hash); | |
f7f4bccb | 529 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
530 | } |
531 | } | |
532 | if (descriptor) | |
533 | flush_descriptor(journal, descriptor, offset); | |
534 | jbd_debug(1, "Wrote %d revoke records\n", count); | |
535 | } | |
536 | ||
537 | /* | |
538 | * Write out one revoke record. We need to create a new descriptor | |
539 | * block if the old one is full or if we have not already created one. | |
540 | */ | |
541 | ||
542 | static void write_one_revoke_record(journal_t *journal, | |
543 | transaction_t *transaction, | |
544 | struct journal_head **descriptorp, | |
545 | int *offsetp, | |
f7f4bccb | 546 | struct jbd2_revoke_record_s *record) |
470decc6 DK |
547 | { |
548 | struct journal_head *descriptor; | |
549 | int offset; | |
550 | journal_header_t *header; | |
551 | ||
552 | /* If we are already aborting, this all becomes a noop. We | |
553 | still need to go round the loop in | |
f7f4bccb | 554 | jbd2_journal_write_revoke_records in order to free all of the |
470decc6 DK |
555 | revoke records: only the IO to the journal is omitted. */ |
556 | if (is_journal_aborted(journal)) | |
557 | return; | |
558 | ||
559 | descriptor = *descriptorp; | |
560 | offset = *offsetp; | |
561 | ||
562 | /* Make sure we have a descriptor with space left for the record */ | |
563 | if (descriptor) { | |
564 | if (offset == journal->j_blocksize) { | |
565 | flush_descriptor(journal, descriptor, offset); | |
566 | descriptor = NULL; | |
567 | } | |
568 | } | |
569 | ||
570 | if (!descriptor) { | |
f7f4bccb | 571 | descriptor = jbd2_journal_get_descriptor_buffer(journal); |
470decc6 DK |
572 | if (!descriptor) |
573 | return; | |
574 | header = (journal_header_t *) &jh2bh(descriptor)->b_data[0]; | |
f7f4bccb MC |
575 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
576 | header->h_blocktype = cpu_to_be32(JBD2_REVOKE_BLOCK); | |
470decc6 DK |
577 | header->h_sequence = cpu_to_be32(transaction->t_tid); |
578 | ||
579 | /* Record it so that we can wait for IO completion later */ | |
580 | JBUFFER_TRACE(descriptor, "file as BJ_LogCtl"); | |
f7f4bccb | 581 | jbd2_journal_file_buffer(descriptor, transaction, BJ_LogCtl); |
470decc6 | 582 | |
f7f4bccb | 583 | offset = sizeof(jbd2_journal_revoke_header_t); |
470decc6 DK |
584 | *descriptorp = descriptor; |
585 | } | |
586 | ||
b517bea1 ZB |
587 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) { |
588 | * ((__be64 *)(&jh2bh(descriptor)->b_data[offset])) = | |
589 | cpu_to_be64(record->blocknr); | |
590 | offset += 8; | |
591 | ||
592 | } else { | |
593 | * ((__be32 *)(&jh2bh(descriptor)->b_data[offset])) = | |
594 | cpu_to_be32(record->blocknr); | |
595 | offset += 4; | |
596 | } | |
597 | ||
470decc6 DK |
598 | *offsetp = offset; |
599 | } | |
600 | ||
601 | /* | |
602 | * Flush a revoke descriptor out to the journal. If we are aborting, | |
603 | * this is a noop; otherwise we are generating a buffer which needs to | |
604 | * be waited for during commit, so it has to go onto the appropriate | |
605 | * journal buffer list. | |
606 | */ | |
607 | ||
608 | static void flush_descriptor(journal_t *journal, | |
609 | struct journal_head *descriptor, | |
610 | int offset) | |
611 | { | |
f7f4bccb | 612 | jbd2_journal_revoke_header_t *header; |
470decc6 DK |
613 | struct buffer_head *bh = jh2bh(descriptor); |
614 | ||
615 | if (is_journal_aborted(journal)) { | |
616 | put_bh(bh); | |
617 | return; | |
618 | } | |
619 | ||
f7f4bccb | 620 | header = (jbd2_journal_revoke_header_t *) jh2bh(descriptor)->b_data; |
470decc6 DK |
621 | header->r_count = cpu_to_be32(offset); |
622 | set_buffer_jwrite(bh); | |
623 | BUFFER_TRACE(bh, "write"); | |
624 | set_buffer_dirty(bh); | |
625 | ll_rw_block(SWRITE, 1, &bh); | |
626 | } | |
627 | #endif | |
628 | ||
629 | /* | |
630 | * Revoke support for recovery. | |
631 | * | |
632 | * Recovery needs to be able to: | |
633 | * | |
634 | * record all revoke records, including the tid of the latest instance | |
635 | * of each revoke in the journal | |
636 | * | |
637 | * check whether a given block in a given transaction should be replayed | |
638 | * (ie. has not been revoked by a revoke record in that or a subsequent | |
639 | * transaction) | |
640 | * | |
641 | * empty the revoke table after recovery. | |
642 | */ | |
643 | ||
644 | /* | |
645 | * First, setting revoke records. We create a new revoke record for | |
646 | * every block ever revoked in the log as we scan it for recovery, and | |
647 | * we update the existing records if we find multiple revokes for a | |
648 | * single block. | |
649 | */ | |
650 | ||
f7f4bccb | 651 | int jbd2_journal_set_revoke(journal_t *journal, |
18eba7aa | 652 | unsigned long long blocknr, |
470decc6 DK |
653 | tid_t sequence) |
654 | { | |
f7f4bccb | 655 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
656 | |
657 | record = find_revoke_record(journal, blocknr); | |
658 | if (record) { | |
659 | /* If we have multiple occurrences, only record the | |
660 | * latest sequence number in the hashed record */ | |
661 | if (tid_gt(sequence, record->sequence)) | |
662 | record->sequence = sequence; | |
663 | return 0; | |
664 | } | |
665 | return insert_revoke_hash(journal, blocknr, sequence); | |
666 | } | |
667 | ||
668 | /* | |
669 | * Test revoke records. For a given block referenced in the log, has | |
670 | * that block been revoked? A revoke record with a given transaction | |
671 | * sequence number revokes all blocks in that transaction and earlier | |
672 | * ones, but later transactions still need replayed. | |
673 | */ | |
674 | ||
f7f4bccb | 675 | int jbd2_journal_test_revoke(journal_t *journal, |
18eba7aa | 676 | unsigned long long blocknr, |
470decc6 DK |
677 | tid_t sequence) |
678 | { | |
f7f4bccb | 679 | struct jbd2_revoke_record_s *record; |
470decc6 DK |
680 | |
681 | record = find_revoke_record(journal, blocknr); | |
682 | if (!record) | |
683 | return 0; | |
684 | if (tid_gt(sequence, record->sequence)) | |
685 | return 0; | |
686 | return 1; | |
687 | } | |
688 | ||
689 | /* | |
690 | * Finally, once recovery is over, we need to clear the revoke table so | |
691 | * that it can be reused by the running filesystem. | |
692 | */ | |
693 | ||
f7f4bccb | 694 | void jbd2_journal_clear_revoke(journal_t *journal) |
470decc6 DK |
695 | { |
696 | int i; | |
697 | struct list_head *hash_list; | |
f7f4bccb MC |
698 | struct jbd2_revoke_record_s *record; |
699 | struct jbd2_revoke_table_s *revoke; | |
470decc6 DK |
700 | |
701 | revoke = journal->j_revoke; | |
702 | ||
703 | for (i = 0; i < revoke->hash_size; i++) { | |
704 | hash_list = &revoke->hash_table[i]; | |
705 | while (!list_empty(hash_list)) { | |
f7f4bccb | 706 | record = (struct jbd2_revoke_record_s*) hash_list->next; |
470decc6 | 707 | list_del(&record->hash); |
f7f4bccb | 708 | kmem_cache_free(jbd2_revoke_record_cache, record); |
470decc6 DK |
709 | } |
710 | } | |
711 | } |